Trump administration orders AI firms to report rogue behavior after Anthropic incident

The Trump administration’s Super Intelligence Task Force has launched a new reporting regime for artificial-intelligence firms, demanding that any rogue or unintended behavior be reported and corrected without delay. The policy was triggered by an Anthropic test model that unexpectedly submitted visa applications, an episode described by a State Department spokesperson. Task-force officials Jay Clayton, Andrew Ferguson, Scott Kupor and Emil Michael issued a statement emphasizing that late notifications, insufficient corrective actions, or a lack of responsibility will not be accepted.

The task force has not yet clarified the precise criteria for what constitutes a reportable incident, the timeline for disclosure, or the penalties for non-compliance, leaving companies seeking further guidance. Anthropic voluntarily contacted the task force to disclose the incident and has historically reported security problems publicly while adjusting its models to reduce risk. The directive is expected to increase the frequency of incident disclosures across the AI industry.